Bug: Resample removes timezone localization from Period.start_time

[ngehrsitz]

Code Sample

from pytz import timezone
index = DatetimeIndex(['2019-01-01 06:00', '2019-01-01 07:00', '2019-01-01 08:00'], tz=timezone('Europe/Berlin'))
s = Series(index=index, data=[1, 2, 3])
print(s.index.tz)
rs = s.resample("3h", kind="period").sum()
for x in rs.index:
    print(x.start_time.tz)
    print(x.end_time.tz)

Problem description

All the Periods in the resampled PeriodIndex should have their start_time and end_time TimeStamps set with the correct Timezone.
Since the the DateTimeIndex was tz aware the attributes in the PeriodIndex should be too.
This is probably related to #13238 and #15777

Expected Output

Europe/Berlin
Europe/Berlin
Europe/Berlin

Tested Version

pandas 0.25.0

[mroeschke]

We normally show a warning when converting a tz aware Timestamp to a Period (applied in #22549)

In [15]: s.index[0].to_period('3h')
/anaconda3/envs/pandas-dev/bin/ipython:1: UserWarning: Converting to Period representation will drop timezone information.
  #!/anaconda3/envs/pandas-dev/bin/python
Out[15]: Period('2019-01-01 06:00', '3H')

Guess in this case the warning is not being surfaced.

[jreback]

slight OT: the kind arg to resample should be deprecated as this is a very confusing api and you should convert to a PI if that is what you want. the implementation currently is very opaque.

[ngehrsitz]

@mroeschke I am not getting any warning. Neither in the Python 3.6 Console nor in PyCharm.
@jreback The reason why i am using resample in the firstplace is that i have a data that is randomly distributed and indexed with a DateTimeIndex. If do resample without the kind argument my index starts at second 12 and not at 18 like i want it to. I also have dynamically changing datapoints and frequencies so i am not sure how to set base correctly.

Code Sample

index = DatetimeIndex(['2019-01-01 06:17:18', '2019-01-01 06:17:22', '2019-01-01 06:18:00'], tz=timezone('US/Eastern'))
d = DataFrame(index=index, data=[[1, 1], [2, 2], [3, 3]])
print(d.index.tz)
incorrect = d.resample("8s").sum()
rs = d.resample("8s", kind="period").sum()
for x in rs.index:
	print(x.start_time.tz)
	print(x.end_time.tz)

If you are not working with your native timezone this requires nasty hacks since you now have to carry the timezone data along with the resampled dataframe.